基于约束HMM的变点检测算法

Change Point Detection Based on Constrained Hidden Markov Model

时间序列的变点分析在现今社会各个领域中都有着广泛的应用.针对时间序列进行变点分析中要求变点状态需要连续持续一定的时间的应用背景,提出了一种结合状态最短连续长度约束的隐马尔可夫模型.给出了约束Baum-Welch训练算法和约束Viterbi状态提取算法.应用在仿真数据和GNP数据集的实验表明,结合状态最短连续长度约束的HMM相比于一般HMM在时间序列变点检测中效率较高.

The change point detection of time series is widely applied in various fields. In some applications, a minimum period is required before a state change. Motivated by such applications, a constrained Hidden Markov Model, which combines with the shortest state continuous length constraint, is proposed in this study. Moreover, a constrained Baum-Welch training algorithm and a constrained Viterbi state extraction algorithm are also given. And experimental results based on the simulation data and GNP data sets indicate that the constrained HMM has higher performance than the general HMM.